cc: Update Main RendererCapabilities on DeferredInitialize

cc/layers/texture_layer.cc

Index: cc/layers/texture_layer.cc
diff --git a/cc/layers/texture_layer.cc b/cc/layers/texture_layer.cc
index bc6b58a0342145bce3130c02d6983591f33855f9..0f683695feb843411c5cfa7a297ec840b25bdec3 100644
--- a/cc/layers/texture_layer.cc
+++ b/cc/layers/texture_layer.cc
@@ -276,6 +276,12 @@ void TextureLayer::PushPropertiesTo(LayerImpl* layer) {
   }
 }
 
+void TextureLayer::OnOutputSurfaceCreated() {
+  if (uses_mailbox_) {
+    holder_ref_.reset();

[boliu, 2014/02/13 20:14:39]

I removed the DCHECK for now. This broke context_lost test on the gpu bots.

The issue here is that in a chrome context loss, it recreates another renderer
with the same RendererCapabilities. So there is nothing stopping a new commit
going through, even if the context is lost.

I still think it's correct here to drop the ref here. But to make the DCHECK
work, we have to pass a bool param saying whether the renderer caps actually
changed with the new OS, and only assert single ref if it did change. Thought
I'd get your opinion first before implementing this though.

[danakj, 2014/02/14 17:59:11]

Maybe we can do this without adding more LayerAPI.. would it work to just
SetNeedsUpdate on the tree when renderer caps change, and then in
Layer::Update() it could decide to drop things as needed. This layer does not
even use the RendererCapabilities, so maybe I'm not sure why it cares that they
change even?

[boliu, 2014/02/14 18:44:42]

RendererCapabilities is used in the
layer_tree_host()->UsingSharedMemoryResources(). I think checking if that
changed in Update works as well. Trying to add a test for it now.

+  }
+}
+
 Region TextureLayer::VisibleContentOpaqueRegion() const {
   if (contents_opaque())
     return visible_content_rect();